光谱、形状特征结合的多精度图像分割算法与应用

由于高分辨率遥感图像的数据量和计算复杂性骤增,影像噪声、光谱混淆现象更为突出,这对传统主要依据于像元光谱特征的图像处理与分析方法提出了极大挑战。为此,特征更为丰富、在噪声处理和知识融合上,更具优势的面向对象的图像分析方法,逐步成为高分辨率遥感图像应用的研究热点。面向对象图像分析的第一步,以及关键一步是图像分割。本文设计了多精度图像分割算法:(1)采用降水分水岭变换进行初步分割获取图像次一级斑块,即分割亚基元;(2)设计一种可重复合并的快速图斑合并方法,进行亚基元的层次归并获得最后分割斑块,完成图像分割。在合并过程中,斑块之间的差异指标是其光谱合并代价、形状合并代价的加权和,合并结束的标志是斑块间两两合并代价超过尺度参数的平方。设置不同的尺度参数,则可实现多精度图像分割过程。实验证明,方法分割效果较好,并在算法效率上满足实际应用需求,可以开展后续图像分类、专题信息提取等工作。

Multi-resolution Remotely Sensed Image Segmentation in Combination with Spectral and Shape Features

The redundant image details of high resolution remotely sensed imagery may cause more distinct spectral diffusion and noise distribution.It sometimes makes traditional pixel-based image analyzing methods inapplicable.Nowadays,object-oriented image analyses have become a hot research topic in the field of remotely sensed image processing and interpretation.It is because the latter has stronger capability to deal with noise,and it can introduce more abundant image features and expertise knowledge in analyses.The first and most important step of object-oriented image analysis is image segmentation,which segments an image into many visual homogenous parcels.Based on these parcels,which are ＂objects＂,not ＂pixels＂,more features can be extracted which facilitate the succeeding image interpretation.In this study,we designed and implemented a multi-resolution image segmentation method combining spectral and shape features with reference to the basic ideas of eCognition,a famous object-oriented image analyzing software package.The algorithm includes the following steps.（1） The initial segmentation parcels,called the ＇sub feature units＇,are obtained with rainfalling watershed algorithm,which is of very fast speed and fair segmentation precision.（2） A fast region merging technique is used to merge these sub feature units in a hierarchy way.Based on the sub feature units,a merge cost function integrating spectral heterogeneity and shape heterogeneity is designed to guide the merging of parcels.The use of the shape features is to make the merged parcels more regular in shapes.A scale parameter is used to control the merging process,which stops a merge when the minimal merging cost of parcels exceeds its power.A multi-resolution segmentation can be implemented by setting different scale parameters,for smaller scales mean less cost while merging which create smaller parcels,and vice versa.Several experiments on high spatial resolution remotely sensed imagery were carried out to validate our method.In most cases,the multi-resolution segmentation produced highly visually homogeneous image objects with arbitrary resolutions on different types of data.